Toward building a fair peer recommender to support help-seeking in online learning

Abstract

Help-seeking is a valuable practice in online discussion forums. However, the asynchronicity and information overload of online discussion forums have made it challenging for help seekers and providers to connect effectively. This study formulated a new method to provide fair and accurate insights toward building a peer recommender to support help-seeking in online learning. Specifically, we developed the fair network embedding (Fair-NE) model and compared it with existing popular models. We trained and evaluated the models with a large dataset consisting of 187,450 discussion post-reply pairs by 10,182 Algebra I online learners from 2015 to 2020. Finally, we examined models with representation fairness, predictive accuracy, and predictive fairness. The results showed that the Fair-NE can achieve superior fairness in genders and races while retaining competitive predictive accuracy. This study marks a paradigm change from previous investigation and evaluation of fair artificial intelligence to proactively build fair artificial intelligence in education.

Keywords:

• fair artificial intelligence
• peer recommenders
• help-seeking
• online discussion forums

Disclosure statement

No potential conflict of interest was declared by the authors.

Funding information

The research reported here was supported by the Institute of Education Sciences, US Department of Education, through Grant R305C160004 to the University of Florida, the University of Florida AI Catalyst Grant -P0195022, and the University of Florida Informatics Institute Seed Funding. The opinions expressed are those of the authors and do not represent the views of the University of Florida, Institute of Education Sciences, or those of the US Department of Education.

Data availability statement

The data that support the findings of this study are available from the corresponding author, WX, upon reasonable request.

Additional information

Funding

Institute for Education Sciences (IES);University of Florida Informatics Institute Seed Grant;

Notes on contributors

Chenglu Li

Chenglu Li is a doctoral student in the Educational Technology Program at the University of Florida. His research interests include learning analytics, Bayesian statistics, fair AI for educational use, educational software development, and educational games.

Wanli Xing

Wanli Xing is an assistant professor of educational technology at the University of Florida. His research interests are artificial intelligence, learning analytics, STEM education, and online learning.

Walter L. Leite

Walter L. Leite is a full professor of research and evaluation methodology at the University of Florida. His research interests are exploring how data mining and machine learning methods may assist in statistical modeling for theory development and causal inference.